Skip to content

Week 1 Slides: What Is This World?

Title Slide

Week 1: What Is This World?

From Chat to Agent

An introduction to LLMs, agents, and the tools we'll use.

Agenda

  • What is an LLM? (And what it is not)
  • What does "agentic" mean?
  • The agent loop
  • Terminal + git essentials
  • Lab overview

What Is an LLM?

Large Language Model

  • A machine learning system trained on vast amounts of text
  • Does one thing: predicts the next token (word/piece of text)
  • Generates answers one token at a time
User: "What is an agent?"
↓
LLM: token 1: "An"
LLM: token 2: "agent"
LLM: token 3: "is"
... (continues)

What an LLM Is NOT

🚫 Not a Search Engine

  • Does not look things up on the internet
  • Generates text from learned patterns
  • Hallucination: sounds confident but is wrong

🚫 Not Always Right

  • Makes mistakes (wrong APIs, outdated info, subtle bugs)
  • Doesn't have real-time knowledge
  • Statistical, not deterministic

🚫 Not Human Reasoning

  • Predicts patterns, not step-by-step logic
  • May struggle with things you find simple
  • May excel at things you find hard

Why We Use LLMs Anyway

LLMs excel at:

✓ Explaining code ✓ Generating boilerplate ✓ Brainstorming solutions ✓ Reviewing code for style/bugs ✓ Writing documentation

Key: Use them wisely. Verify output. Don't trust hallucinations.

Chat vs. Agent

Chat Interface (e.g., ChatGPT)

You: "Write a reverse function"
↓
ChatGPT: "Here's code..."
↓
You: Copy → Paste → Run → Test manually

You do the work.

Chat vs. Agent (cont.)

Agentic System (e.g., OpenCode)

You: "Add a reverse function"
↓
Agent: Reads your code & structure
↓
Agent: Writes function in the right place
↓
Agent: Runs tests automatically
↓
Agent: Observes errors → Fixes them
↓
You: See the result. It's done.

The agent does the work. You supervise.

The Agent Loop

Five Steps, Repeating

  1. Read — Inspect files, run commands
  2. Think — LLM decides what to do
  3. Act — Execute code, write files
  4. Observe — See the results
  5. Loop — Repeat until done

The Agent Loop (Visual)

┌──────────────┐
│  User Request│
└──────┬───────┘
       │
       ▼
   ┌────────┐
   │  READ  │
   └───┬────┘
       │
       ▼
   ┌────────┐
   │ THINK  │
   └───┬────┘
       │
       ▼
   ┌────────┐
   │  ACT   │
   └───┬────┘
       │
       ▼
   ┌────────┐
   │OBSERVE │
   └───┬────┘
       │
       └──→ Loop or Done?

Terminal Essentials

Five Commands

Command What It Does
pwd Show current directory
ls List files & folders
cd Change directory
cat Print file contents
echo Print text

Git Essentials

Three Operations

  1. Clone — Copy a repo to your computer 

    git clone https://github.com/example/repo.git

  2. Status — See what you've changed 

    git status

  3. Commit — Save a snapshot 

    git add .
git commit -m "Fixed bug X"

This Week's Labs

Lab 1.1: Terminal + Git

  • Install Terminal (macOS) or WSL (Windows)
  • Clone a starter repo
  • Navigate with cd, ls
  • Make your first commit

Lab 1.2: Prompting

  • Write 5 prompts to a free chat LLM
  • Rate them using a rubric
  • Learn what makes a good prompt

Reflection Questions

By the end of today, you should be able to answer:

  1. What does an LLM actually do? (Token prediction, not reasoning.)
  2. Why might it give you wrong information confidently?
  3. What is the agent loop, in your own words?
  4. How is "agent" different from "chat"?
  5. When would you use the terminal instead of clicking?

What's Next

Week 2: Installing & First Run

  • Install OpenCode
  • Set up a model provider
  • Run your first agentic task
  • Learn plan mode vs. build mode

See you next week!